The Application for Object Distance Estimation and Collision Warning

The world of machine learning especially that of deep learning, for the past 10 years has been an area of high interest for many computer scientists which see in this field tremendous opportunity to bring forward effective solutions to problems which were very difficult if not impossible to do with traditional software development methods. In many sectors today machine learning has proven to be an essential tool in the optimisation of IT products, and we can see that in various sectors such Robotics, Security, Medicine, Self-driving car etc. In this project we want to propose a solution based on deep learning to the problem of road accidents which is now the 9th cause of death globally and the first cause of death among the young people aged 5-29 years. The report [1] by WHO on road safety in 2018 states that this number is steadily increasing, with less develop countries (especially in Africa and South East Asia) as being the most affected. This same report says that if nothing is done road accident will be the 5th cause of death among the youths by 2030. Despite the number of measures elaborated in this report to reduce this number, we firmly believe this deep learning project can significantly reduce this number. A research conducted by Frank et al[1] shows that 6 out of 10 front end car crashes could be avoided if we had safety systems which reacted less than a second earlier than the driver. Given the great improvement in deep learning frameworks and availability of data and high computational machines, there is no barrier preventing us to take advantage of these to propose a solution. For this reason, we propose a project which will detect objects on our roads, estimate the distance of these object from the camera and alert the driver if this distance is equal or less than the threshold value(15meters). Our project aims to assist the driver and alert him as soon as possible in order for him to take appropriate actions as soon as possible which can avoid any collision or significantly reduce the impact significantly. We plan to use state of the arts object detection models like YOLO to identify the target object classes and use depth maps from monocular camera to be give an accurate estimate of the distance of the detected object from the camera. Since this system will be use in a delicate environment, one major requirement of this system is the real-time behaviour and a high accuracy for the detected and estimated distance. A second requirement is to make the system cheap and easy useable comparatively to the other existing methods. That is why we decided to use monocular camera images and depth maps which makes the solution cheap and innovative. This project (prototype) provide room for bigger and more complete project which will contribute to the creation of tool which can save lives and improve security on our roads. Key Words: deep learning, distance estimation, collision avoidance, depth map, machine learning